Letta

## At a Glance:

Letta (formerly MemGPT) is an open-source agent framework for building AI with advanced memory that supports local execution via CLI, API integration through Python and TypeScript SDKs, and features such as skills, subagents, and model-agnostic design.

## Overview:

Letta is an agent framework designed to build AI applications with persistent memory and self-improvement capabilities. It operates through two primary channels: a local CLI tool (`letta-code`) for running agents directly in the terminal and a stateful agents API with SDKs for integration into custom applications. The system is structured around skills and subagents, including pre-built options for advanced memory and continual learning. Letta remains model-agnostic, allowing users to choose from a range of large language models based on their performance needs.

## Key Decision Points:

* **Local vs. API-first development**: The `letta-code` CLI runs agents entirely locally on a user’s computer, while the Letta API is designed for embedding stateful agents into external applications.
* **Agent architecture**: Users can extend functionality through a skills system and delegate tasks using subagents, with the framework bundling pre-built components specifically for memory and learning.
* **Model flexibility**: The framework is model-agnostic and does not lock users into a specific provider, though guidance is provided on which models yield the best performance.

## Core Features:

* **Local agent runtime**: Enables the execution of memory-enabled agents directly in a terminal using a single CLI command.
* **Stateful agents API**: Provides a dedicated API for building applications with agents that maintain state over time.
* **Python and TypeScript SDKs**: Offers software development kits for integrating agent capabilities into applications written in these languages.
* **Skills and subagents**: Supports a modular architecture where agents can be enhanced with specific skills or managed through subordinate agent processes.
* **Continual learning system**: Includes bundled components that enable agents to learn and improve their behavior during operation.

## Use Cases:

* Developers who want to run a persistent AI agent locally to assist with coding and computer-based tasks from the terminal.
* Application developers looking to integrate stateful, memory-equipped AI agents into their own software via an API and SDKs.

## Open-Source Alternative Value:

As an open-source framework, Letta provides direct access to the codebase underlying its agent memory and self-improving architecture. Developers can run agents locally through the `letta-code` CLI without dependency on a hosted service for execution. The framework’s model-agnostic design and its SDKs for Python and TypeScript allow for customization and integration into self-defined environments based on the available source code and APIs.